• A novel approach to estimate total plant carbon, accounting for below-ground carbon • Developed a new biomass-to-carbon conversion factor tailored for industrial hemp • Proposed a fusion of LiDAR and hyperspectral data to estimate total plant carbon • Knowledge distillation improved estimation for above and below-ground biomass Industrial hemp ( Cannabis sativa L.) is emerging as a compelling crop for climate change mitigation and economic development, given its rapid growth, substantial biomass yield, significant carbon sequestration capacity, and diverse industrial applications. Estimating total plant carbon (TPC), particularly the below-ground biomass (BGB) component, is challenging due to the complexity of capturing root biomass. To address this challenge, this study presents a methodology for estimating TPC, encompassing both above-ground biomass (AGB) and BGB, by integrating UAV-based Light Detection and Ranging (LiDAR) and hyperspectral data. We derived structural metrics from LiDAR, indicative of canopy complexity, and salient hyperspectral bands, reflective of plant biochemistry. In situ, destructive sampling provided AGB and BGB measurements, while laboratory analyses provided %C values that served as a conversion factor for biomass estimates. We used these datasets to validate the accuracy of the remote sensing-based models. We applied two distinct modeling approaches: statistical machine learning models and a knowledge distillation framework utilizing a teacher-student paradigm. Under our initial modeling approach, the SVR model achieved R² values of 0.817 for AGB and 0.869 for BGB, which improved to 0.915 and 0.926, respectively, under the knowledge distillation framework. Strong correlations (R > 0.93) between estimated AGB and BGB were observed, reflecting the inherent linear trend in root-shoot interrelation. Moreover, TPC varied substantially across study areas, reaching 6.176 kg/plot. Our findings highlight the advancement of using remote sensing to estimate TPC, offering a more comprehensive and accurate assessment than previous methods, which often overlooked BGB.
Pawar et al. (Sun,) studied this question.